Table 3.

Test result data on TRUS Datasets. Each neural network has two rows of conclusion data. The first row of data corresponds to the original neural network framework proposed by the authors. In contrast, the second row of data represents the test results after incorporating our method SAA-SDM

Name	mDice	mIoU	Accuracy	Precision	Recall	Kappa	Acceleration
UNet	0.9523	0.9089	0.9751	0.9503	0.9543	0.9355	-
	0.9493	0.9035	0.9734	0.9418	0.9573	0.9313	42.86%
SegNet	0.6499	0.4813	0.7981	0.592	0.7203	0.5099	-
	0.9272	0.8643	0.9616	0.9154	0.9393	0.9012	∞
Attention UNet	0.9413	0.91372	0.9766	0.9677	0.9862	0.9391	-
	0.9358	0.90581	0.9743	0.9681	0.9860	0.9332	23.81%
UNet +  +	0.9078	0.8312	0.9503	0.8762	0.9418	0.8738	-
	0.9346	0.8773	0.9650	0.9089	0.9619	0.9108	47.06%
SegFormer	0.9127	0.8083	0.9431	0.8734	0.9159	0.8546	-
	0.9456	0.8681	0.9632	0.9302	0.9562	0.9045	∞
SegFormer (pretrained)	0.9583	0.9200	0.9783	0.9583	0.9584	0.9436	-
	0.9581	0.9195	0.9783	0.9622	0.9539	0.9434	25%
SegMenter	0.5289	0.3595	0.7386	0.4979	0.5641	0.3409	-
	0.8812	0.7876	0.9377	0.8747	0.8878	0.8390	∞
SegMenter (pretrained)	0.9473	0.8999	0.9725	0.9431	0.9516	0.9287	-
	0.9485	0.9021	0.9730	0.9418	0.9554	0.9303	0%

Bolded characters indicate accuracy data where the method in this paper has significantly improved compared to the original neural network